AI usecases in Finance
(Third in a Series)
Artificial Intelligence (AI) presents us with the opportunity to protect investors, provide transparency about investment decisions, decrease costs, increase fairness, reduce risk, and streamline the entire process. It has been a bumpy road to get here (just ask Knight Capital about a 2012 $440 million loss due to trading errors with their new trading software), but we’re on the verge of massive changes because our technology is now sufficiently capable.
There are several aspects of computing technology that have come together at precisely the right time for us to be able to see the beginning of true AI. The first of those things is the sheer capability of modern computers regarding both speed and processing power.
The Need For Speed
The CDC 6600 (Control Data Corporation) was the first Supercomputer, and was the fastest computer on the planet from 1964-69 until it was displaced by its successor, the CDC 7600. The 6600 could manage a stunning 3 MegaFLOPS (3 million or 3 × 106 Floating Point Operations per second).
Nowadays your household desktop computer can manage GigaFLOPS (109) or TeraFLOPS (1012) with relative ease. In 1961, one GFLOP of processing power would have cost $145 billion (US 2013 dollars). By 1984, that was down to $43 million per GFLOP. In 1997 it reduced to just $42,000, and by 2003 it was just $100. By 2012 it had fallen below one dollar to just 73¢, but by the end of 2013 was down to 12¢ per GFLOP. In October 2017, it reached just 3¢ per GFLOP. Computing is now fast and cheap.
The Next Thing
Computer RAM and storage has become inexpensive and fast. These have had to get bigger and faster so they can provide enough information to the CPU (Central Processing Unit) to keep it busy. All the speed in the world is useless if you can’t get sufficiently large amounts of data to the CPU and then have a place to put it once the work is done. Without all these elements working together, it would be like trying to fill a goldfish bowl with a firehose.
Speak To Me
The ability of computers to interpret ordinary speech is a major accomplishment. This is called Natural Language Programming (NLP, distinct from the silly and now debunked 1970s theory of Neuro Linguistic Programming).
You may have become accustomed to speaking to Alexa, Siri, Lyra, or any number of other digital personal assistants. Right now there is a generation growing up that may look back on today thinking “People used to type words? Why? Couldn’t they just tell the computer what they wanted?”
Right now, this is still funny to many of us, but soon people will wonder what this whole 33-second movie scene was all about…
Finance is all about calculations. The earliest computer originated between 250-70 B.C.E., known as the Antikythera mechanism which was used to calculate astronomical positions, dates, eclipses, and much more, to an astonishingly high degree of accuracy using only handmade gears. For an analog device, operated with a crank that was turned by hand, there was nothing comparable. Like our modern computers, it handled incredibly complex mathematical calculations for the user, even if they had absolutely no idea about how it was accomplished.
AI gives us even more options. Early computers had fixed inputs and fixed outputs. The Antikythera device was amazing but only had one output for one input. Artificial intelligence, on the other hand, is a system that has been provided with a basic set of facts, and instructions on how to manipulate and interpret those facts, to create “learning.”
Such an AI system would exist within an isolated learning environment where it would combine those facts in billions of different ways. It wouldn’t get bored, no matter how small the change. It will then determine if the output signifies an advance or improvement according to the parameter’s given by the programmer.
After billions or trillions of operations, it ends up with a list of strategies that improved the situation, worsened it, or contributed nothing. This is how humans learn, too, although we’re much quicker at isolating bad strategies (such as placing a hand on a hot stove element), because of our feedback mechanisms. Once you add a few layers of complexity, the AI could determine answers faster than a human.
Stock trading is complex because there are so many elements affecting how it functions—it behaves as if it were a living, breathing entity. An AI system provided with the entire history of worldwide trading probably would not be able to make infallible predictions based on that information alone.
Provide it with the knowledge of what caused trading events, and it could then make connections between those events and their impact on the market. Processing and validating news on a moment-to-moment basis could be done in mere seconds, and the AI would extrapolate the most likely thing that would occur next and invest to take advantage of it. To someone unaware of its computer nature, it would appear to be the world’s best “instinctive trader.”
401(k)’s are probably the worst invention of 1978. They have helped a lot of people put away some money for retirement, but it forced each person to become a stock market investor without any preparation or skill. Some employers help by matching contributions, but each plan is different, with very specific choices available or, more like an IRA, the ability to choose just about any investment.
While many people have made significant increases in their retirement savings, usually relying on outside “expert” advice, few have made monstrous fortunes; many more have lost much of their savings. Selecting where to put money is often a guessing game for the average citizen.
The problem is that many people don’t know what they’re doing. So-called expert advice might be ill-informed opinion or, in some cases, specifically designed to manipulate the market to the advantage of that “advisor.”
AI Financial Advisors
An AI, armed with your personal information such as income, risk tolerance, age, and desired goals, could provide a dozen different strategies from which you could choose. This could all be accomplished through a ChatBot interface which can be entirely text-based for record-keeping but accessed by voice.
Alternatively, with the current state of CGI, it would be utterly trivial to animate a fairly human face so that expressions, mouth movements, and other identifying characteristics would mimic a human being. It would be much more relatable than simply reading the text. Clients could engage in “human” conversations, complete with regional accents and speech patterns, which would be accurately interpreted by the AI.
The difference is that the responses are not fixed. Depending on who asks the questions, the answers would be unique for that person. Moreover, with the client’s permission, the AI could continue to track the chosen goals and recommend changes to portfolios or investment decisions to keep the plan on track.
They might even go so far as to let the AI control the investment for the client for a trivial fee. Now the client doesn’t have to be an investment expert, and the 401(k) becomes more generally useful to the average citizen. Again, it’s not limited to 401(k) plans; it could just as easily manage an IRA or a plain old-fashioned stock and bond portfolio.
AI Supports Dealing Rooms
ChatBots are great and make customers feel supported and valued, but AI has a significant contribution to make in the dealing rooms, too. Meshing Artificial Intelligence with Big Data is almost perfect synergy. AIs are better than humans at analysis and finding interconnections; humans are (for the time being) far superior in judging the mood of the market, or the quality of the team at the management reins of an up-and-coming business that has the potential to grow substantially in the future.
Turning the AI loose on your Big Data means that it can make available nearly invisible trends that are too obscure for a human to recognize. Combined with human intuition and knowledge, it provides an immense amount of support in the decision-making process.
Although we’re only at the beginning of the Artificial Intelligence revolution, and its capabilities are still somewhat limited, all of that is going to change everything from Finance to Genetics; from Banking to new Drug Development; from Architecture to Art. Just ask Ken Jennings the former Jeopardy champion, who was beaten by IBM’s Watson Supercomputer.
Watson was designed, by the best available IBM engineers, to play Jeopardy, and that alone. They threw millions of dollars and years of research at the project to beat the world’s best human, and they succeeded, but what they were left with was not a true AI. It was a Jeopardy-question Interpreter with thousands and thousands of co-processors and a terabyte of memory. It really couldn’t do much else—by the very definition—a one-trick pony.
In succeeding years they have increased its capabilities, and it is now available online. Anybody can use its AI capabilities for experimental or development work. But don’t worry about the rise of Skynet (the AI from the Terminator movies that attempt to destroy humanity).
In his TED Talk, Shyam Sankar describes a future where the strengths of humans and the strengths of AI computing are used cooperatively. In one empirical example, he shows that a Grand Master chess champion, aided by a laptop, could beat the most powerful supercomputer. The surprise result was that a Grand Master aided by a supercomputer was beaten by a team of two amateur players using three comparatively weak laptops, much like the one you probably own.
The Grand Master’s knowledge and the supercomputer’s superior computational power were beaten by the amateurs’ ability to delve deeply into each position and analyze them in incredible detail. AIs can’t run the world successfully on their own because they lack insight. Humans may be able to manage that, but it will be incredibly slow because we think slowly.
The future has a spot for everybody, but it will be a spot where we all cooperate to achieve goals. Our future will be a good place to be.